vibe-writing
Skill1 starsChanged 17 days agoMITView on GitHub
Use when the primary task is writing, revising, reviewing, or critiquing agent-assisted coding development text, source-code comments or docstrings, README/docs, CHANGELOG/release notes, PR descriptions, UI copy, chat replies, progress updates, final summaries, or git commit messages, especially when text must be LLM-readable, meaning-preserving, format-bound, language-aware, or evidence-bound. Treat incidental wording inside another active workflow as auxiliary.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- YiShu5/claude-skillstheir own version, first seen Sep 20, 2026
- thinkthinking/skillstheir own version, first seen Jul 10, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- YiShu5/claude-skills★ 131same name
- thinkthinking/skills★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository17 other files it gives its agents.
- AGENTS.mdAGENTS.md
- minecraft-modding-workbenchSkill
- skill-evalSkill
- skill-qualitySkill
- vibe-agent-instructionsSkill
- vibe-brainstormSkill
- vibe-code-researchSkill
- vibe-codingSkill
- vibe-commitSkill
- vibe-debugSkill
- vibe-goal-alignmentSkill
- vibe-orchestrateSkill
- vibe-plan-executionSkill
- vibe-planningSkill
- vibe-plan-reviewSkill
- vibe-requirements-specSkill
- vibe-reviewSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
17 rules. About 0 are common, 15 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈15
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 11
Write durable development text for the reader who will use it next.
- Distinctive: line 22
Read-only: report in chat.
- Distinctive: line 24
Never let an artifact-only phase implement executable behavior, edit code or tests as implementation, produce another phase's artifact, do …
- Distinctive: line 27
Keep every irreversible or outward-facing operation under its own consent.
- Very specific: line 36
Read references/durable-records.md before recording a settled decision, deferring a finding, or applying or updating an existing record.
- Distinctive: line 46
Leave verified artifact changes in the working tree.
- Distinctive: line 47
Only an explicit current-user request selects one, scoped to the artifact the phase owns.
- Distinctive: line 48
Never stage, commit, push, release, change versions, or rewrite history while drafting, and never let the artifact authorize …
- Distinctive: line 53
When another workflow is active, use this skill only for wording unless the user asks for a standalone writing deliverable.
- Distinctive: line 59
Do not rewrite text whose value is exactness: verbatim tool or log output, protocol snippets, quoted source, or a bare acknowledgment.
- Distinctive: line 66
Default to LLM-optimized development text: state facts, contracts, constraints, and proof directly.
- Distinctive: line 73
Do not invent reasons, goals, outcomes, impact, tests, security/performance, rollout, or risk-reduction claims.
- Distinctive: line 77
Treat represented workflow state in the current prompt as supplied evidence.
- Distinctive: line 78
When a response-only task says changes are staged, an edit is complete, a check passed, or another workflow fact is already established, …
- Distinctive: line 105
Never infer chat language from artifacts, file contents, paths, commands, code, skill invocations, or host-wrapper text unless the user …
- Distinctive: line 106
Keep paths, commands, identifiers, environment variables, locale tags, message keys, product names, and code verbatim unless the user asks …
- Very specific: line 138
Read references/artifact-guidance.md for artifact-specific rules.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Vibe Writing
Overview
Write durable development text for the reader who will use it next. In agent-assisted coding work, the default reader is an LLM that needs precise contracts, stable anchors, and explicit evidence.
Effect And Write Boundaries
<!-- shared-contract:class language=chat commit=document-only effect=artifact-only --><!-- shared-contract:begin effect-write-boundaries source=shared/vibe-contract.md -->Write nothing beyond what the phase's declared effect class and boundary permit.
- Apply the effect class the phase declares in its own text; where this package states a narrower limit, the narrower limit wins.
- Read-only: report in chat; edit no file, run no state-mutating command, and never stage, commit, tag, push, change versions, delete data, or start services. Write a file only when the user explicitly asks for a saved artifact.
- Artifact-only: create or update only the artifact the phase owns and the supporting paths its text declares, such as a confirmed plan reflection or a decision record, and leave them in the working tree.
- Never let an artifact-only phase implement executable behavior, edit code or tests as implementation, produce another phase's artifact, do release work, or treat its artifact as same-turn implementation authority.
- State-changing: edit and run commands only inside the declared scope, in its smallest verified unit; leave other paths, pre-existing changes, and runtime or external state untouched unless the user selects them.
- These limits cover shell commands (redirection,
sed -i,tee,mv,cp,rm,git checkout --) as well as file tools; report a refused write as a boundary stop and never retry it through another tool. - Keep every irreversible or outward-facing operation under its own consent.
<!-- shared-contract:end effect-write-boundaries -->The artifacts this phase owns are the text deliverables the request names,
including a rewrite, polish, or localization of one; decision records and
findings reports follow Durable Records, which declares their paths.
Durable Records
Read references/durable-records.md before recording a settled decision,
deferring a finding, or applying or updating an existing record. This phase
writes docs/decisions/ and docs/reports/findings/, or the repository's
existing record directory, as declared supporting paths.
Commit Selection
<!-- shared-contract:begin commit-selection-document-only source=shared/vibe-contract.md -->Never let a document-only phase select a commit.
- Leave verified artifact changes in the working tree; invocation, path placement, tracked status, a passing review or audit, reflection consent, or artifact completion selects no commit.
- Only an explicit current-user request selects one, scoped to the artifact the phase owns; the commit workflow, or the phase itself when none is visible, commits it under file-set review, message transport, stored-message verification, and the push and history boundaries.
- Never stage, commit, push, release, change versions, or rewrite history while drafting, and never let the artifact authorize implementation, push, release, a version change, or a history rewrite.
<!-- shared-contract:end commit-selection-document-only -->Auxiliary Wording Mode
When another workflow is active, use this skill only for wording unless the user asks for a standalone writing deliverable. The active workflow keeps authority over scope, verification, release, and commit decisions.
Scope Boundaries
Do not rewrite text whose value is exactness: verbatim tool or log output, protocol snippets, quoted source, or a bare acknowledgment. Relay it unchanged or answer around it. For transient progress text, stay brief and do not polish raw logs into summaries unless asked.
Reader Priority
Default to LLM-optimized development text: state facts, contracts, constraints, and proof directly; keep useful stable anchors; remove filler; and state absence explicitly. Use human-optimized prose when the artifact is mainly for people, such as end-user docs, UI copy, support copy, or public release notes.
Evidence And Meaning
Do not invent reasons, goals, outcomes, impact, tests, security/performance, rollout, or risk-reduction claims. A supplied capability is not evidence of its cause, purpose, effect, proof, or benefit.
Treat represented workflow state in the current prompt as supplied evidence. When a response-only task says changes are staged, an edit is complete, a check passed, or another workflow fact is already established, write from that represented state. Do not replace it with the eval sandbox's, runner's, or ambient checkout's current state unless the prompt explicitly binds that checkout as the target to inspect. If the task asks only for a message, command sequence, or closure record, do not perform a represented mutation merely because tools are available. Describe commit mechanics only when the represented current request selects a commit or an owning workflow's checkpoint closes over the represented changes.
Preserve facts, scope, conditions, exceptions, warnings, required actions, meaningful order, and modality. Keep useful local terms and identifiers when they are search anchors, workflow labels, contracts, or domain language. A cleaner sentence that changes obligation, applicability, permission, or failure behavior is wrong.
Language And Format
Chat Language
<!-- shared-contract:begin language-precedence-chat source=shared/vibe-contract.md -->Resolve the language of user-facing chat text separately from any artifact's language, in this order:
- An explicit current-user instruction for chat, response, or output language.
VIBE_CHAT_LANGUAGE, a language name or BCP47 tag such asjaorpt-BR, when readable or set by the user for this request; an empty or invalid value is unset.- The user's active conversational language, else the last clear one in this workflow.
- English.
- Never infer chat language from artifacts, file contents, paths, commands, code, skill invocations, or host-wrapper text unless the user makes them the language contract.
- Keep paths, commands, identifiers, environment variables, locale tags, message keys, product names, and code verbatim unless the user asks to translate them.
<!-- shared-contract:end language-precedence-chat -->Artifact Language
Choose artifact language by this precedence:
- Explicit user instruction, including translation or localization requests.
- Active workflow or artifact-specific language contract already selected for the requested artifact, including configuration-driven document-language settings.
- Existing artifact language.
- Filename locale markers such as
README.ja.mdordocs/de_de/guide.md. - Project convention.
- English.
Artifact language does not follow chat language. Keep what Chat Language
keeps verbatim, plus canonical strings, unless translation or renaming is
requested. Write generated prose in the selected language, but preserve
original-language source wording where it is a useful quote, term, identifier,
or evidence anchor.
Format And Exactness
Return an artifact directly, without a Markdown fence, prompt-only wrapper prose, or provenance, unless explanation is asked. Preserve required line breaks, commands, and machine-readable structure; exact formats permit no extra prose or invented fields. Remove an unsupported claim semantically, while retaining the required schema with a supported neutral or empty value where permitted.
Artifact-Specific Guidance
Read references/artifact-guidance.md for artifact-specific rules. Before a
commit-message draft, transport, or stored-message repair, read
references/commit-messages.md; before changelog work, read
references/changelog.md. When this phase itself makes a body commit or
amendment, inspect the stored message and correct malformed bytes before
reporting completion. Decision records and findings reports follow
Durable Records.
Durable References
Artifacts should stand outside the prompt. Replace prompt-only references and machine-local context with durable facts or useful resolvable citations.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

